Method

Every 30 minutes during market hours the strategy rebuilds its universe, scores it, and checks the positions it already holds. Three times a day, mid-morning, midday and late session, it also opens a decision window in which it may enter something new.

Exits are checked on every cycle, not just in decision windows. A position that breaks its rule does not wait for a convenient time to be closed.

Candidates are scored on three things:

• Trend: is price above its short, medium and long moving averages.
• Momentum: a blend of recent returns across several lookbacks, so a single hot week cannot dominate.
• Volume: is today's volume confirming the move, measured against the 20-day average.

The top few names by composite score go forward for a closer look. Ranking highly is necessary but nowhere near sufficient.

A trade is rejected outright if any of these fail. They are checked mechanically, after any judgement has been applied, and they always win:

• The name must be on the approved candidate list for that cycle.
• No position may exceed 10% of the account.
• At most 4 positions open at once.
• A daily loss of 2% halts all new buying for the day.
• Buying power must actually cover the order.
• The market must genuinely be open.
• Long positions only, no shorting, no options, no leverage.
• Overheated names are rejected: a momentum reading above 85 is a late entry, not an opportunity.
• Volume below 0.8× its 20-day average is rejected: no confirmation, no trade.

Momentum strategies bleed in falling markets, so entries are gated on the state of the broad market in three tiers:

• Healthy: the market is above its medium-term trend. Full position sizing.
• Pullback: below the medium-term trend but the long-term uptrend is intact. Entries continue at half size, and only the cleanest setups qualify.
• Defensive: below the long-term trend. No new positions at all.

Exits keep running normally in every tier. The strategy will stop buying long before it stops protecting what it holds.

Every position is sized so that being stopped out costs roughly the same fraction of the account: about 1%, regardless of the name. Volatile stocks get fewer shares, quiet ones get more.

This means each trade is an equal risk bet, not an equal dollarbet, which is what makes the trades on the track record comparable to one another. The 10% cap still applies on top and can only ever reduce the size.

A position is closed when any of the following fires:

• Protective stop: a hard stop below entry, scaled to the name's own volatility. It rests at the broker as a live order, so it works overnight and at weekends whether or not anything is running.
• Trailing stop: the stop ratchets up as the position gains, and never down. It tightens once a trade is in profit.
• Trend break: price closes below its 20-day average and the reason for owning it has gone.
• Time stop: a position that has gone nowhere for long enough frees its slot for something that might.
• Overheated: a parabolic run is taken as an exit signal, not a reason to hold on.

A language model reviews the ranked shortlist and writes the plain-English reasoning published on this site. It is consultative only. It cannot place an order, cannot size a position, and cannot relax a single gate above. Everything it suggests is re-checked mechanically and rejected if it fails.

It also has no live news access. It reasons from price, volume and trend, and it is explicitly instructed not to speculate about events it cannot see.

In parallel, a rules-only baseline records what a purely mechanical version would have done each cycle. That comparison is logged continuously. It is how the question “does the judgement layer add anything at all?” gets answered with data rather than opinion.

It is simulated. A paper brokerage account, no real money, no real orders. Simulated fills do not model slippage, partial fills, or the liquidity limits a real order would meet.

It is not advice.This is a published record of what one automated strategy did and why. It is not a recommendation, and it is not tailored to anyone's circumstances.

The sample is small. Judging any strategy on a few dozen trades is statistically meaningless. That is precisely why the whole record is published rather than a highlight reel, so the sample size is visible instead of implied.